Neural networks for the H1 experiment

With its rising luminosity the HERA-2 period does not only produce more physics events for the H1 detector but also - and in a more dramatic way higher backgrounds. The neural network trigger becomes more important than before because background rejection is now needed even for triggers that had no rate problem before. We will discuss several examples of newly developed neural network triggers ranging from deeply virtual Compton scattering to charged current interactions. But also in offline analysis neural networks are a very powerful tool to separate event classes by complex pattern recognition. We will discuss the example of the search for instantons in the high Q² regime and show that the neural networks give higher separation powers for the event sample purification than other techniques.